skulldc

OpenBot is an MIT-licensed workspace where each AI coworker gets its own computer

CopilotKit's OpenBot gives agents a browser, files and tools under sign-in, permissions, approvals and audit, but relies on a separately licensed Intelligence service for memory and learning.

What it is

OpenBot is a TypeScript project from CopilotKit, the company behind the AG-UI agent protocol. The README describes it as a modular, self-hostable workspace for AI coworkers. The repository description is more concrete: each AI coworker gets a computer of its own, with a browser, files and tools, and every action is decided before it happens and recorded after.

The repository was created on 2026-08-17. At the time of writing it showed 11490 stars, 291 of them gained in the last day on GitHub trending, and 1425 forks; treat these as snapshot figures. The application template is under the MIT license, and the README says it can be forked, modified and white-labeled.

How it is put together

The README describes three parts:

  • Frontends for Mac, Windows, web, iOS, Android, Slack and Teams, all connected to the same workspace. Conversations, interactive results, a view of the bot’s computer and approvals adapt to each surface.
  • The OpenBot backend, which runs the workspace API and built-in Bots, connects external agents and tools, manages Bot computers, and enforces identity, permissions, action policies and audit. It deploys on Kubernetes with PostgreSQL and persistent storage.
  • CopilotKit Intelligence, which provides durable conversations, memory, Automatic Learning and governance services.

Agents connect through AG-UI and tools through MCP. Built-in Bots run inside the OpenBot server, while existing agents can stay where they already run. The README lists Amazon Bedrock AgentCore, Claude Managed Agents, LangChain’s Managed Deep Agents and Microsoft Foundry Agent Service among the options, alongside your own infrastructure. Models can be OpenAI, Claude, Gemini, or open models behind an OpenAI-compatible endpoint. Computer execution is pluggable: the default is a Kubernetes setup, or a compatible external computer provider can be connected.

The licensing catch

Only the application template is MIT. CopilotKit Intelligence is separately licensed. According to the README, personal use on a laptop needs a free local Intelligence license, which desktop setup handles; a Kubernetes evaluation needs a free trial license; and Kubernetes production use needs a production license. OpenBot for Business includes Intelligence for use with OpenBot.

Intelligence can run on your own machine or infrastructure. The README says fully local operation is possible with a compatible local model, but initial downloads, account sign-in and local license renewal still need internet access, and a cloud model or connected online app still sends requests to that service.

Getting it running

For laptops there is a notarized .dmg for Apple silicon and Intel Macs and a signed installer for Windows x64. The app combines the OpenBot frontend and a Docker-based backend. For a cluster, the README gives a Helm install that deploys OpenBot and Intelligence together on AWS EKS, Azure AKS, Google GKE or another Kubernetes cluster with amd64 nodes and a default StorageClass:

helm dependency build charts/openbot
helm install openbot charts/openbot \
 --namespace openbot --create-namespace \
 --values charts/openbot/ci/vanilla-values.yaml \
 --values openbot-secrets.yaml \
 --timeout 15m

The evaluation is opened through a local port-forward on port 3001. The README warns that this configuration is single-user and that anyone who can reach it is an administrator, so sign-in and TLS should be configured before a deployment is shared. Running from source needs Docker, Bun 1.4+ (some tests fail on 1.3), a CopilotKit Intelligence project and license, and a model key.


Source: CopilotKit/OpenBot on GitHub